1 // SPDX-License-Identifier: GPL-2.0
4 * Copyright (C) 1992, 1993 Krishna Balasubramanian
5 * Many improvements/fixes by Bruno Haible.
6 * Replaced `struct shm_desc' by `struct vm_area_struct', July 1994.
7 * Fixed the shm swap deallocation (shm_unuse()), August 1998 Andrea Arcangeli.
9 * /proc/sysvipc/shm support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
10 * BIGMEM support, Andrea Arcangeli <andrea@suse.de>
11 * SMP thread shm, Jean-Luc Boyard <jean-luc.boyard@siemens.fr>
12 * HIGHMEM support, Ingo Molnar <mingo@redhat.com>
13 * Make shmmax, shmall, shmmni sysctl'able, Christoph Rohland <cr@sap.com>
14 * Shared /dev/zero support, Kanoj Sarcar <kanoj@sgi.com>
15 * Move the mm functionality over to mm/shmem.c, Christoph Rohland <cr@sap.com>
17 * support for audit of ipc object properties and permission changes
18 * Dustin Kirkland <dustin.kirkland@us.ibm.com>
22 * Pavel Emelianov <xemul@openvz.org>
24 * Better ipc lock (kern_ipc_perm.lock) handling
25 * Davidlohr Bueso <davidlohr.bueso@hp.com>, June 2013.
28 #include <linux/slab.h>
30 #include <linux/hugetlb.h>
31 #include <linux/shm.h>
32 #include <linux/init.h>
33 #include <linux/file.h>
34 #include <linux/mman.h>
35 #include <linux/shmem_fs.h>
36 #include <linux/security.h>
37 #include <linux/syscalls.h>
38 #include <linux/audit.h>
39 #include <linux/capability.h>
40 #include <linux/ptrace.h>
41 #include <linux/seq_file.h>
42 #include <linux/rwsem.h>
43 #include <linux/nsproxy.h>
44 #include <linux/mount.h>
45 #include <linux/ipc_namespace.h>
46 #include <linux/rhashtable.h>
48 #include <linux/uaccess.h>
52 struct shmid_kernel
/* private to the kernel */
54 struct kern_ipc_perm shm_perm
;
55 struct file
*shm_file
;
56 unsigned long shm_nattch
;
57 unsigned long shm_segsz
;
61 struct pid
*shm_cprid
;
62 struct pid
*shm_lprid
;
63 struct ucounts
*mlock_ucounts
;
66 * The task created the shm object, for
67 * task_lock(shp->shm_creator)
69 struct task_struct
*shm_creator
;
72 * List by creator. task_lock(->shm_creator) required for read/write.
73 * If list_empty(), then the creator is dead already.
75 struct list_head shm_clist
;
76 struct ipc_namespace
*ns
;
79 /* shm_mode upper byte flags */
80 #define SHM_DEST 01000 /* segment will be destroyed on last detach */
81 #define SHM_LOCKED 02000 /* segment will not be swapped */
83 struct shm_file_data
{
85 struct ipc_namespace
*ns
;
87 const struct vm_operations_struct
*vm_ops
;
90 #define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
92 static const struct file_operations shm_file_operations
;
93 static const struct vm_operations_struct shm_vm_ops
;
95 #define shm_ids(ns) ((ns)->ids[IPC_SHM_IDS])
97 #define shm_unlock(shp) \
98 ipc_unlock(&(shp)->shm_perm)
100 static int newseg(struct ipc_namespace
*, struct ipc_params
*);
101 static void shm_open(struct vm_area_struct
*vma
);
102 static void shm_close(struct vm_area_struct
*vma
);
103 static void shm_destroy(struct ipc_namespace
*ns
, struct shmid_kernel
*shp
);
104 #ifdef CONFIG_PROC_FS
105 static int sysvipc_shm_proc_show(struct seq_file
*s
, void *it
);
108 void shm_init_ns(struct ipc_namespace
*ns
)
110 ns
->shm_ctlmax
= SHMMAX
;
111 ns
->shm_ctlall
= SHMALL
;
112 ns
->shm_ctlmni
= SHMMNI
;
113 ns
->shm_rmid_forced
= 0;
115 ipc_init_ids(&shm_ids(ns
));
119 * Called with shm_ids.rwsem (writer) and the shp structure locked.
120 * Only shm_ids.rwsem remains locked on exit.
122 static void do_shm_rmid(struct ipc_namespace
*ns
, struct kern_ipc_perm
*ipcp
)
124 struct shmid_kernel
*shp
;
126 shp
= container_of(ipcp
, struct shmid_kernel
, shm_perm
);
127 WARN_ON(ns
!= shp
->ns
);
129 if (shp
->shm_nattch
) {
130 shp
->shm_perm
.mode
|= SHM_DEST
;
131 /* Do not find it any more */
132 ipc_set_key_private(&shm_ids(ns
), &shp
->shm_perm
);
135 shm_destroy(ns
, shp
);
139 void shm_exit_ns(struct ipc_namespace
*ns
)
141 free_ipcs(ns
, &shm_ids(ns
), do_shm_rmid
);
142 idr_destroy(&ns
->ids
[IPC_SHM_IDS
].ipcs_idr
);
143 rhashtable_destroy(&ns
->ids
[IPC_SHM_IDS
].key_ht
);
147 static int __init
ipc_ns_init(void)
149 shm_init_ns(&init_ipc_ns
);
153 pure_initcall(ipc_ns_init
);
155 void __init
shm_init(void)
157 ipc_init_proc_interface("sysvipc/shm",
158 #if BITS_PER_LONG <= 32
159 " key shmid perms size cpid lpid nattch uid gid cuid cgid atime dtime ctime rss swap\n",
161 " key shmid perms size cpid lpid nattch uid gid cuid cgid atime dtime ctime rss swap\n",
163 IPC_SHM_IDS
, sysvipc_shm_proc_show
);
166 static inline struct shmid_kernel
*shm_obtain_object(struct ipc_namespace
*ns
, int id
)
168 struct kern_ipc_perm
*ipcp
= ipc_obtain_object_idr(&shm_ids(ns
), id
);
171 return ERR_CAST(ipcp
);
173 return container_of(ipcp
, struct shmid_kernel
, shm_perm
);
176 static inline struct shmid_kernel
*shm_obtain_object_check(struct ipc_namespace
*ns
, int id
)
178 struct kern_ipc_perm
*ipcp
= ipc_obtain_object_check(&shm_ids(ns
), id
);
181 return ERR_CAST(ipcp
);
183 return container_of(ipcp
, struct shmid_kernel
, shm_perm
);
187 * shm_lock_(check_) routines are called in the paths where the rwsem
188 * is not necessarily held.
190 static inline struct shmid_kernel
*shm_lock(struct ipc_namespace
*ns
, int id
)
192 struct kern_ipc_perm
*ipcp
;
195 ipcp
= ipc_obtain_object_idr(&shm_ids(ns
), id
);
199 ipc_lock_object(ipcp
);
201 * ipc_rmid() may have already freed the ID while ipc_lock_object()
202 * was spinning: here verify that the structure is still valid.
203 * Upon races with RMID, return -EIDRM, thus indicating that
204 * the ID points to a removed identifier.
206 if (ipc_valid_object(ipcp
)) {
207 /* return a locked ipc object upon success */
208 return container_of(ipcp
, struct shmid_kernel
, shm_perm
);
211 ipc_unlock_object(ipcp
);
212 ipcp
= ERR_PTR(-EIDRM
);
216 * Callers of shm_lock() must validate the status of the returned ipc
217 * object pointer and error out as appropriate.
219 return ERR_CAST(ipcp
);
222 static inline void shm_lock_by_ptr(struct shmid_kernel
*ipcp
)
225 ipc_lock_object(&ipcp
->shm_perm
);
228 static void shm_rcu_free(struct rcu_head
*head
)
230 struct kern_ipc_perm
*ptr
= container_of(head
, struct kern_ipc_perm
,
232 struct shmid_kernel
*shp
= container_of(ptr
, struct shmid_kernel
,
234 security_shm_free(&shp
->shm_perm
);
239 * It has to be called with shp locked.
240 * It must be called before ipc_rmid()
242 static inline void shm_clist_rm(struct shmid_kernel
*shp
)
244 struct task_struct
*creator
;
246 /* ensure that shm_creator does not disappear */
250 * A concurrent exit_shm may do a list_del_init() as well.
251 * Just do nothing if exit_shm already did the work
253 if (!list_empty(&shp
->shm_clist
)) {
255 * shp->shm_creator is guaranteed to be valid *only*
256 * if shp->shm_clist is not empty.
258 creator
= shp
->shm_creator
;
262 * list_del_init() is a nop if the entry was already removed
265 list_del_init(&shp
->shm_clist
);
266 task_unlock(creator
);
271 static inline void shm_rmid(struct shmid_kernel
*s
)
274 ipc_rmid(&shm_ids(s
->ns
), &s
->shm_perm
);
278 static int __shm_open(struct shm_file_data
*sfd
)
280 struct shmid_kernel
*shp
;
282 shp
= shm_lock(sfd
->ns
, sfd
->id
);
287 if (shp
->shm_file
!= sfd
->file
) {
293 shp
->shm_atim
= ktime_get_real_seconds();
294 ipc_update_pid(&shp
->shm_lprid
, task_tgid(current
));
300 /* This is called by fork, once for every shm attach. */
301 static void shm_open(struct vm_area_struct
*vma
)
303 struct file
*file
= vma
->vm_file
;
304 struct shm_file_data
*sfd
= shm_file_data(file
);
307 /* Always call underlying open if present */
308 if (sfd
->vm_ops
->open
)
309 sfd
->vm_ops
->open(vma
);
311 err
= __shm_open(sfd
);
313 * We raced in the idr lookup or with shm_destroy().
314 * Either way, the ID is busted.
320 * shm_destroy - free the struct shmid_kernel
323 * @shp: struct to free
325 * It has to be called with shp and shm_ids.rwsem (writer) locked,
326 * but returns with shp unlocked and freed.
328 static void shm_destroy(struct ipc_namespace
*ns
, struct shmid_kernel
*shp
)
330 struct file
*shm_file
;
332 shm_file
= shp
->shm_file
;
333 shp
->shm_file
= NULL
;
334 ns
->shm_tot
-= (shp
->shm_segsz
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
337 if (!is_file_hugepages(shm_file
))
338 shmem_lock(shm_file
, 0, shp
->mlock_ucounts
);
340 ipc_update_pid(&shp
->shm_cprid
, NULL
);
341 ipc_update_pid(&shp
->shm_lprid
, NULL
);
342 ipc_rcu_putref(&shp
->shm_perm
, shm_rcu_free
);
346 * shm_may_destroy - identifies whether shm segment should be destroyed now
348 * Returns true if and only if there are no active users of the segment and
349 * one of the following is true:
351 * 1) shmctl(id, IPC_RMID, NULL) was called for this shp
353 * 2) sysctl kernel.shm_rmid_forced is set to 1.
355 static bool shm_may_destroy(struct shmid_kernel
*shp
)
357 return (shp
->shm_nattch
== 0) &&
358 (shp
->ns
->shm_rmid_forced
||
359 (shp
->shm_perm
.mode
& SHM_DEST
));
363 * remove the attach descriptor vma.
364 * free memory for segment if it is marked destroyed.
365 * The descriptor has already been removed from the current->mm->mmap list
366 * and will later be kfree()d.
368 static void __shm_close(struct shm_file_data
*sfd
)
370 struct shmid_kernel
*shp
;
371 struct ipc_namespace
*ns
= sfd
->ns
;
373 down_write(&shm_ids(ns
).rwsem
);
374 /* remove from the list of attaches of the shm segment */
375 shp
= shm_lock(ns
, sfd
->id
);
378 * We raced in the idr lookup or with shm_destroy().
379 * Either way, the ID is busted.
381 if (WARN_ON_ONCE(IS_ERR(shp
)))
382 goto done
; /* no-op */
384 ipc_update_pid(&shp
->shm_lprid
, task_tgid(current
));
385 shp
->shm_dtim
= ktime_get_real_seconds();
387 if (shm_may_destroy(shp
))
388 shm_destroy(ns
, shp
);
392 up_write(&shm_ids(ns
).rwsem
);
395 static void shm_close(struct vm_area_struct
*vma
)
397 struct file
*file
= vma
->vm_file
;
398 struct shm_file_data
*sfd
= shm_file_data(file
);
400 /* Always call underlying close if present */
401 if (sfd
->vm_ops
->close
)
402 sfd
->vm_ops
->close(vma
);
407 /* Called with ns->shm_ids(ns).rwsem locked */
408 static int shm_try_destroy_orphaned(int id
, void *p
, void *data
)
410 struct ipc_namespace
*ns
= data
;
411 struct kern_ipc_perm
*ipcp
= p
;
412 struct shmid_kernel
*shp
= container_of(ipcp
, struct shmid_kernel
, shm_perm
);
415 * We want to destroy segments without users and with already
416 * exit'ed originating process.
418 * As shp->* are changed under rwsem, it's safe to skip shp locking.
420 if (!list_empty(&shp
->shm_clist
))
423 if (shm_may_destroy(shp
)) {
424 shm_lock_by_ptr(shp
);
425 shm_destroy(ns
, shp
);
430 void shm_destroy_orphaned(struct ipc_namespace
*ns
)
432 down_write(&shm_ids(ns
).rwsem
);
433 if (shm_ids(ns
).in_use
)
434 idr_for_each(&shm_ids(ns
).ipcs_idr
, &shm_try_destroy_orphaned
, ns
);
435 up_write(&shm_ids(ns
).rwsem
);
438 /* Locking assumes this will only be called with task == current */
439 void exit_shm(struct task_struct
*task
)
442 struct shmid_kernel
*shp
;
443 struct ipc_namespace
*ns
;
447 if (list_empty(&task
->sysvshm
.shm_clist
)) {
452 shp
= list_first_entry(&task
->sysvshm
.shm_clist
, struct shmid_kernel
,
456 * 1) Get pointer to the ipc namespace. It is worth to say
457 * that this pointer is guaranteed to be valid because
458 * shp lifetime is always shorter than namespace lifetime
459 * in which shp lives.
460 * We taken task_lock it means that shp won't be freed.
465 * 2) If kernel.shm_rmid_forced is not set then only keep track of
466 * which shmids are orphaned, so that a later set of the sysctl
469 if (!ns
->shm_rmid_forced
)
470 goto unlink_continue
;
473 * 3) get a reference to the namespace.
474 * The refcount could be already 0. If it is 0, then
475 * the shm objects will be free by free_ipc_work().
477 ns
= get_ipc_ns_not_zero(ns
);
480 list_del_init(&shp
->shm_clist
);
486 * 4) get a reference to shp.
487 * This cannot fail: shm_clist_rm() is called before
488 * ipc_rmid(), thus the refcount cannot be 0.
490 WARN_ON(!ipc_rcu_getref(&shp
->shm_perm
));
493 * 5) unlink the shm segment from the list of segments
494 * created by current.
495 * This must be done last. After unlinking,
496 * only the refcounts obtained above prevent IPC_RMID
497 * from destroying the segment or the namespace.
499 list_del_init(&shp
->shm_clist
);
504 * 6) we have all references
505 * Thus lock & if needed destroy shp.
507 down_write(&shm_ids(ns
).rwsem
);
508 shm_lock_by_ptr(shp
);
510 * rcu_read_lock was implicitly taken in shm_lock_by_ptr, it's
511 * safe to call ipc_rcu_putref here
513 ipc_rcu_putref(&shp
->shm_perm
, shm_rcu_free
);
515 if (ipc_valid_object(&shp
->shm_perm
)) {
516 if (shm_may_destroy(shp
))
517 shm_destroy(ns
, shp
);
522 * Someone else deleted the shp from namespace
523 * idr/kht while we have waited.
524 * Just unlock and continue.
529 up_write(&shm_ids(ns
).rwsem
);
530 put_ipc_ns(ns
); /* paired with get_ipc_ns_not_zero */
534 static vm_fault_t
shm_fault(struct vm_fault
*vmf
)
536 struct file
*file
= vmf
->vma
->vm_file
;
537 struct shm_file_data
*sfd
= shm_file_data(file
);
539 return sfd
->vm_ops
->fault(vmf
);
542 static int shm_may_split(struct vm_area_struct
*vma
, unsigned long addr
)
544 struct file
*file
= vma
->vm_file
;
545 struct shm_file_data
*sfd
= shm_file_data(file
);
547 if (sfd
->vm_ops
->may_split
)
548 return sfd
->vm_ops
->may_split(vma
, addr
);
553 static unsigned long shm_pagesize(struct vm_area_struct
*vma
)
555 struct file
*file
= vma
->vm_file
;
556 struct shm_file_data
*sfd
= shm_file_data(file
);
558 if (sfd
->vm_ops
->pagesize
)
559 return sfd
->vm_ops
->pagesize(vma
);
565 static int shm_set_policy(struct vm_area_struct
*vma
, struct mempolicy
*new)
567 struct file
*file
= vma
->vm_file
;
568 struct shm_file_data
*sfd
= shm_file_data(file
);
571 if (sfd
->vm_ops
->set_policy
)
572 err
= sfd
->vm_ops
->set_policy(vma
, new);
576 static struct mempolicy
*shm_get_policy(struct vm_area_struct
*vma
,
579 struct file
*file
= vma
->vm_file
;
580 struct shm_file_data
*sfd
= shm_file_data(file
);
581 struct mempolicy
*pol
= NULL
;
583 if (sfd
->vm_ops
->get_policy
)
584 pol
= sfd
->vm_ops
->get_policy(vma
, addr
);
585 else if (vma
->vm_policy
)
586 pol
= vma
->vm_policy
;
592 static int shm_mmap(struct file
*file
, struct vm_area_struct
*vma
)
594 struct shm_file_data
*sfd
= shm_file_data(file
);
598 * In case of remap_file_pages() emulation, the file can represent an
599 * IPC ID that was removed, and possibly even reused by another shm
600 * segment already. Propagate this case as an error to caller.
602 ret
= __shm_open(sfd
);
606 ret
= call_mmap(sfd
->file
, vma
);
611 sfd
->vm_ops
= vma
->vm_ops
;
613 WARN_ON(!sfd
->vm_ops
->fault
);
615 vma
->vm_ops
= &shm_vm_ops
;
619 static int shm_release(struct inode
*ino
, struct file
*file
)
621 struct shm_file_data
*sfd
= shm_file_data(file
);
625 shm_file_data(file
) = NULL
;
630 static int shm_fsync(struct file
*file
, loff_t start
, loff_t end
, int datasync
)
632 struct shm_file_data
*sfd
= shm_file_data(file
);
634 if (!sfd
->file
->f_op
->fsync
)
636 return sfd
->file
->f_op
->fsync(sfd
->file
, start
, end
, datasync
);
639 static long shm_fallocate(struct file
*file
, int mode
, loff_t offset
,
642 struct shm_file_data
*sfd
= shm_file_data(file
);
644 if (!sfd
->file
->f_op
->fallocate
)
646 return sfd
->file
->f_op
->fallocate(file
, mode
, offset
, len
);
649 static unsigned long shm_get_unmapped_area(struct file
*file
,
650 unsigned long addr
, unsigned long len
, unsigned long pgoff
,
653 struct shm_file_data
*sfd
= shm_file_data(file
);
655 return sfd
->file
->f_op
->get_unmapped_area(sfd
->file
, addr
, len
,
659 static const struct file_operations shm_file_operations
= {
662 .release
= shm_release
,
663 .get_unmapped_area
= shm_get_unmapped_area
,
664 .llseek
= noop_llseek
,
665 .fallocate
= shm_fallocate
,
669 * shm_file_operations_huge is now identical to shm_file_operations,
670 * but we keep it distinct for the sake of is_file_shm_hugepages().
672 static const struct file_operations shm_file_operations_huge
= {
675 .release
= shm_release
,
676 .get_unmapped_area
= shm_get_unmapped_area
,
677 .llseek
= noop_llseek
,
678 .fallocate
= shm_fallocate
,
681 bool is_file_shm_hugepages(struct file
*file
)
683 return file
->f_op
== &shm_file_operations_huge
;
686 static const struct vm_operations_struct shm_vm_ops
= {
687 .open
= shm_open
, /* callback for a new vm-area open */
688 .close
= shm_close
, /* callback for when the vm-area is released */
690 .may_split
= shm_may_split
,
691 .pagesize
= shm_pagesize
,
692 #if defined(CONFIG_NUMA)
693 .set_policy
= shm_set_policy
,
694 .get_policy
= shm_get_policy
,
699 * newseg - Create a new shared memory segment
701 * @params: ptr to the structure that contains key, size and shmflg
703 * Called with shm_ids.rwsem held as a writer.
705 static int newseg(struct ipc_namespace
*ns
, struct ipc_params
*params
)
707 key_t key
= params
->key
;
708 int shmflg
= params
->flg
;
709 size_t size
= params
->u
.size
;
711 struct shmid_kernel
*shp
;
712 size_t numpages
= (size
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
715 vm_flags_t acctflag
= 0;
717 if (size
< SHMMIN
|| size
> ns
->shm_ctlmax
)
720 if (numpages
<< PAGE_SHIFT
< size
)
723 if (ns
->shm_tot
+ numpages
< ns
->shm_tot
||
724 ns
->shm_tot
+ numpages
> ns
->shm_ctlall
)
727 shp
= kmalloc(sizeof(*shp
), GFP_KERNEL_ACCOUNT
);
731 shp
->shm_perm
.key
= key
;
732 shp
->shm_perm
.mode
= (shmflg
& S_IRWXUGO
);
733 shp
->mlock_ucounts
= NULL
;
735 shp
->shm_perm
.security
= NULL
;
736 error
= security_shm_alloc(&shp
->shm_perm
);
742 sprintf(name
, "SYSV%08x", key
);
743 if (shmflg
& SHM_HUGETLB
) {
747 hs
= hstate_sizelog((shmflg
>> SHM_HUGE_SHIFT
) & SHM_HUGE_MASK
);
752 hugesize
= ALIGN(size
, huge_page_size(hs
));
754 /* hugetlb_file_setup applies strict accounting */
755 if (shmflg
& SHM_NORESERVE
)
756 acctflag
= VM_NORESERVE
;
757 file
= hugetlb_file_setup(name
, hugesize
, acctflag
,
758 HUGETLB_SHMFS_INODE
, (shmflg
>> SHM_HUGE_SHIFT
) & SHM_HUGE_MASK
);
761 * Do not allow no accounting for OVERCOMMIT_NEVER, even
764 if ((shmflg
& SHM_NORESERVE
) &&
765 sysctl_overcommit_memory
!= OVERCOMMIT_NEVER
)
766 acctflag
= VM_NORESERVE
;
767 file
= shmem_kernel_file_setup(name
, size
, acctflag
);
769 error
= PTR_ERR(file
);
773 shp
->shm_cprid
= get_pid(task_tgid(current
));
774 shp
->shm_lprid
= NULL
;
775 shp
->shm_atim
= shp
->shm_dtim
= 0;
776 shp
->shm_ctim
= ktime_get_real_seconds();
777 shp
->shm_segsz
= size
;
779 shp
->shm_file
= file
;
780 shp
->shm_creator
= current
;
782 /* ipc_addid() locks shp upon success. */
783 error
= ipc_addid(&shm_ids(ns
), &shp
->shm_perm
, ns
->shm_ctlmni
);
790 list_add(&shp
->shm_clist
, ¤t
->sysvshm
.shm_clist
);
791 task_unlock(current
);
794 * shmid gets reported as "inode#" in /proc/pid/maps.
795 * proc-ps tools use this. Changing this will break them.
797 file_inode(file
)->i_ino
= shp
->shm_perm
.id
;
799 ns
->shm_tot
+= numpages
;
800 error
= shp
->shm_perm
.id
;
802 ipc_unlock_object(&shp
->shm_perm
);
807 ipc_update_pid(&shp
->shm_cprid
, NULL
);
808 ipc_update_pid(&shp
->shm_lprid
, NULL
);
810 ipc_rcu_putref(&shp
->shm_perm
, shm_rcu_free
);
813 call_rcu(&shp
->shm_perm
.rcu
, shm_rcu_free
);
818 * Called with shm_ids.rwsem and ipcp locked.
820 static int shm_more_checks(struct kern_ipc_perm
*ipcp
, struct ipc_params
*params
)
822 struct shmid_kernel
*shp
;
824 shp
= container_of(ipcp
, struct shmid_kernel
, shm_perm
);
825 if (shp
->shm_segsz
< params
->u
.size
)
831 long ksys_shmget(key_t key
, size_t size
, int shmflg
)
833 struct ipc_namespace
*ns
;
834 static const struct ipc_ops shm_ops
= {
836 .associate
= security_shm_associate
,
837 .more_checks
= shm_more_checks
,
839 struct ipc_params shm_params
;
841 ns
= current
->nsproxy
->ipc_ns
;
843 shm_params
.key
= key
;
844 shm_params
.flg
= shmflg
;
845 shm_params
.u
.size
= size
;
847 return ipcget(ns
, &shm_ids(ns
), &shm_ops
, &shm_params
);
850 SYSCALL_DEFINE3(shmget
, key_t
, key
, size_t, size
, int, shmflg
)
852 return ksys_shmget(key
, size
, shmflg
);
855 static inline unsigned long copy_shmid_to_user(void __user
*buf
, struct shmid64_ds
*in
, int version
)
859 return copy_to_user(buf
, in
, sizeof(*in
));
864 memset(&out
, 0, sizeof(out
));
865 ipc64_perm_to_ipc_perm(&in
->shm_perm
, &out
.shm_perm
);
866 out
.shm_segsz
= in
->shm_segsz
;
867 out
.shm_atime
= in
->shm_atime
;
868 out
.shm_dtime
= in
->shm_dtime
;
869 out
.shm_ctime
= in
->shm_ctime
;
870 out
.shm_cpid
= in
->shm_cpid
;
871 out
.shm_lpid
= in
->shm_lpid
;
872 out
.shm_nattch
= in
->shm_nattch
;
874 return copy_to_user(buf
, &out
, sizeof(out
));
881 static inline unsigned long
882 copy_shmid_from_user(struct shmid64_ds
*out
, void __user
*buf
, int version
)
886 if (copy_from_user(out
, buf
, sizeof(*out
)))
891 struct shmid_ds tbuf_old
;
893 if (copy_from_user(&tbuf_old
, buf
, sizeof(tbuf_old
)))
896 out
->shm_perm
.uid
= tbuf_old
.shm_perm
.uid
;
897 out
->shm_perm
.gid
= tbuf_old
.shm_perm
.gid
;
898 out
->shm_perm
.mode
= tbuf_old
.shm_perm
.mode
;
907 static inline unsigned long copy_shminfo_to_user(void __user
*buf
, struct shminfo64
*in
, int version
)
911 return copy_to_user(buf
, in
, sizeof(*in
));
916 if (in
->shmmax
> INT_MAX
)
917 out
.shmmax
= INT_MAX
;
919 out
.shmmax
= (int)in
->shmmax
;
921 out
.shmmin
= in
->shmmin
;
922 out
.shmmni
= in
->shmmni
;
923 out
.shmseg
= in
->shmseg
;
924 out
.shmall
= in
->shmall
;
926 return copy_to_user(buf
, &out
, sizeof(out
));
934 * Calculate and add used RSS and swap pages of a shm.
935 * Called with shm_ids.rwsem held as a reader
937 static void shm_add_rss_swap(struct shmid_kernel
*shp
,
938 unsigned long *rss_add
, unsigned long *swp_add
)
942 inode
= file_inode(shp
->shm_file
);
944 if (is_file_hugepages(shp
->shm_file
)) {
945 struct address_space
*mapping
= inode
->i_mapping
;
946 struct hstate
*h
= hstate_file(shp
->shm_file
);
947 *rss_add
+= pages_per_huge_page(h
) * mapping
->nrpages
;
950 struct shmem_inode_info
*info
= SHMEM_I(inode
);
952 spin_lock_irq(&info
->lock
);
953 *rss_add
+= inode
->i_mapping
->nrpages
;
954 *swp_add
+= info
->swapped
;
955 spin_unlock_irq(&info
->lock
);
957 *rss_add
+= inode
->i_mapping
->nrpages
;
963 * Called with shm_ids.rwsem held as a reader
965 static void shm_get_stat(struct ipc_namespace
*ns
, unsigned long *rss
,
974 in_use
= shm_ids(ns
).in_use
;
976 for (total
= 0, next_id
= 0; total
< in_use
; next_id
++) {
977 struct kern_ipc_perm
*ipc
;
978 struct shmid_kernel
*shp
;
980 ipc
= idr_find(&shm_ids(ns
).ipcs_idr
, next_id
);
983 shp
= container_of(ipc
, struct shmid_kernel
, shm_perm
);
985 shm_add_rss_swap(shp
, rss
, swp
);
992 * This function handles some shmctl commands which require the rwsem
993 * to be held in write mode.
994 * NOTE: no locks must be held, the rwsem is taken inside this function.
996 static int shmctl_down(struct ipc_namespace
*ns
, int shmid
, int cmd
,
997 struct shmid64_ds
*shmid64
)
999 struct kern_ipc_perm
*ipcp
;
1000 struct shmid_kernel
*shp
;
1003 down_write(&shm_ids(ns
).rwsem
);
1006 ipcp
= ipcctl_obtain_check(ns
, &shm_ids(ns
), shmid
, cmd
,
1007 &shmid64
->shm_perm
, 0);
1009 err
= PTR_ERR(ipcp
);
1013 shp
= container_of(ipcp
, struct shmid_kernel
, shm_perm
);
1015 err
= security_shm_shmctl(&shp
->shm_perm
, cmd
);
1021 ipc_lock_object(&shp
->shm_perm
);
1022 /* do_shm_rmid unlocks the ipc object and rcu */
1023 do_shm_rmid(ns
, ipcp
);
1026 ipc_lock_object(&shp
->shm_perm
);
1027 err
= ipc_update_perm(&shmid64
->shm_perm
, ipcp
);
1030 shp
->shm_ctim
= ktime_get_real_seconds();
1038 ipc_unlock_object(&shp
->shm_perm
);
1042 up_write(&shm_ids(ns
).rwsem
);
1046 static int shmctl_ipc_info(struct ipc_namespace
*ns
,
1047 struct shminfo64
*shminfo
)
1049 int err
= security_shm_shmctl(NULL
, IPC_INFO
);
1051 memset(shminfo
, 0, sizeof(*shminfo
));
1052 shminfo
->shmmni
= shminfo
->shmseg
= ns
->shm_ctlmni
;
1053 shminfo
->shmmax
= ns
->shm_ctlmax
;
1054 shminfo
->shmall
= ns
->shm_ctlall
;
1055 shminfo
->shmmin
= SHMMIN
;
1056 down_read(&shm_ids(ns
).rwsem
);
1057 err
= ipc_get_maxidx(&shm_ids(ns
));
1058 up_read(&shm_ids(ns
).rwsem
);
1065 static int shmctl_shm_info(struct ipc_namespace
*ns
,
1066 struct shm_info
*shm_info
)
1068 int err
= security_shm_shmctl(NULL
, SHM_INFO
);
1070 memset(shm_info
, 0, sizeof(*shm_info
));
1071 down_read(&shm_ids(ns
).rwsem
);
1072 shm_info
->used_ids
= shm_ids(ns
).in_use
;
1073 shm_get_stat(ns
, &shm_info
->shm_rss
, &shm_info
->shm_swp
);
1074 shm_info
->shm_tot
= ns
->shm_tot
;
1075 shm_info
->swap_attempts
= 0;
1076 shm_info
->swap_successes
= 0;
1077 err
= ipc_get_maxidx(&shm_ids(ns
));
1078 up_read(&shm_ids(ns
).rwsem
);
1085 static int shmctl_stat(struct ipc_namespace
*ns
, int shmid
,
1086 int cmd
, struct shmid64_ds
*tbuf
)
1088 struct shmid_kernel
*shp
;
1091 memset(tbuf
, 0, sizeof(*tbuf
));
1094 if (cmd
== SHM_STAT
|| cmd
== SHM_STAT_ANY
) {
1095 shp
= shm_obtain_object(ns
, shmid
);
1100 } else { /* IPC_STAT */
1101 shp
= shm_obtain_object_check(ns
, shmid
);
1109 * Semantically SHM_STAT_ANY ought to be identical to
1110 * that functionality provided by the /proc/sysvipc/
1111 * interface. As such, only audit these calls and
1112 * do not do traditional S_IRUGO permission checks on
1115 if (cmd
== SHM_STAT_ANY
)
1116 audit_ipc_obj(&shp
->shm_perm
);
1119 if (ipcperms(ns
, &shp
->shm_perm
, S_IRUGO
))
1123 err
= security_shm_shmctl(&shp
->shm_perm
, cmd
);
1127 ipc_lock_object(&shp
->shm_perm
);
1129 if (!ipc_valid_object(&shp
->shm_perm
)) {
1130 ipc_unlock_object(&shp
->shm_perm
);
1135 kernel_to_ipc64_perm(&shp
->shm_perm
, &tbuf
->shm_perm
);
1136 tbuf
->shm_segsz
= shp
->shm_segsz
;
1137 tbuf
->shm_atime
= shp
->shm_atim
;
1138 tbuf
->shm_dtime
= shp
->shm_dtim
;
1139 tbuf
->shm_ctime
= shp
->shm_ctim
;
1140 #ifndef CONFIG_64BIT
1141 tbuf
->shm_atime_high
= shp
->shm_atim
>> 32;
1142 tbuf
->shm_dtime_high
= shp
->shm_dtim
>> 32;
1143 tbuf
->shm_ctime_high
= shp
->shm_ctim
>> 32;
1145 tbuf
->shm_cpid
= pid_vnr(shp
->shm_cprid
);
1146 tbuf
->shm_lpid
= pid_vnr(shp
->shm_lprid
);
1147 tbuf
->shm_nattch
= shp
->shm_nattch
;
1149 if (cmd
== IPC_STAT
) {
1151 * As defined in SUS:
1152 * Return 0 on success
1157 * SHM_STAT and SHM_STAT_ANY (both Linux specific)
1158 * Return the full id, including the sequence number
1160 err
= shp
->shm_perm
.id
;
1163 ipc_unlock_object(&shp
->shm_perm
);
1169 static int shmctl_do_lock(struct ipc_namespace
*ns
, int shmid
, int cmd
)
1171 struct shmid_kernel
*shp
;
1172 struct file
*shm_file
;
1176 shp
= shm_obtain_object_check(ns
, shmid
);
1182 audit_ipc_obj(&(shp
->shm_perm
));
1183 err
= security_shm_shmctl(&shp
->shm_perm
, cmd
);
1187 ipc_lock_object(&shp
->shm_perm
);
1189 /* check if shm_destroy() is tearing down shp */
1190 if (!ipc_valid_object(&shp
->shm_perm
)) {
1195 if (!ns_capable(ns
->user_ns
, CAP_IPC_LOCK
)) {
1196 kuid_t euid
= current_euid();
1198 if (!uid_eq(euid
, shp
->shm_perm
.uid
) &&
1199 !uid_eq(euid
, shp
->shm_perm
.cuid
)) {
1203 if (cmd
== SHM_LOCK
&& !rlimit(RLIMIT_MEMLOCK
)) {
1209 shm_file
= shp
->shm_file
;
1210 if (is_file_hugepages(shm_file
))
1213 if (cmd
== SHM_LOCK
) {
1214 struct ucounts
*ucounts
= current_ucounts();
1216 err
= shmem_lock(shm_file
, 1, ucounts
);
1217 if (!err
&& !(shp
->shm_perm
.mode
& SHM_LOCKED
)) {
1218 shp
->shm_perm
.mode
|= SHM_LOCKED
;
1219 shp
->mlock_ucounts
= ucounts
;
1225 if (!(shp
->shm_perm
.mode
& SHM_LOCKED
))
1227 shmem_lock(shm_file
, 0, shp
->mlock_ucounts
);
1228 shp
->shm_perm
.mode
&= ~SHM_LOCKED
;
1229 shp
->mlock_ucounts
= NULL
;
1231 ipc_unlock_object(&shp
->shm_perm
);
1233 shmem_unlock_mapping(shm_file
->f_mapping
);
1239 ipc_unlock_object(&shp
->shm_perm
);
1245 static long ksys_shmctl(int shmid
, int cmd
, struct shmid_ds __user
*buf
, int version
)
1248 struct ipc_namespace
*ns
;
1249 struct shmid64_ds sem64
;
1251 if (cmd
< 0 || shmid
< 0)
1254 ns
= current
->nsproxy
->ipc_ns
;
1258 struct shminfo64 shminfo
;
1259 err
= shmctl_ipc_info(ns
, &shminfo
);
1262 if (copy_shminfo_to_user(buf
, &shminfo
, version
))
1267 struct shm_info shm_info
;
1268 err
= shmctl_shm_info(ns
, &shm_info
);
1271 if (copy_to_user(buf
, &shm_info
, sizeof(shm_info
)))
1278 err
= shmctl_stat(ns
, shmid
, cmd
, &sem64
);
1281 if (copy_shmid_to_user(buf
, &sem64
, version
))
1286 if (copy_shmid_from_user(&sem64
, buf
, version
))
1290 return shmctl_down(ns
, shmid
, cmd
, &sem64
);
1293 return shmctl_do_lock(ns
, shmid
, cmd
);
1299 SYSCALL_DEFINE3(shmctl
, int, shmid
, int, cmd
, struct shmid_ds __user
*, buf
)
1301 return ksys_shmctl(shmid
, cmd
, buf
, IPC_64
);
1304 #ifdef CONFIG_ARCH_WANT_IPC_PARSE_VERSION
1305 long ksys_old_shmctl(int shmid
, int cmd
, struct shmid_ds __user
*buf
)
1307 int version
= ipc_parse_version(&cmd
);
1309 return ksys_shmctl(shmid
, cmd
, buf
, version
);
1312 SYSCALL_DEFINE3(old_shmctl
, int, shmid
, int, cmd
, struct shmid_ds __user
*, buf
)
1314 return ksys_old_shmctl(shmid
, cmd
, buf
);
1318 #ifdef CONFIG_COMPAT
1320 struct compat_shmid_ds
{
1321 struct compat_ipc_perm shm_perm
;
1323 old_time32_t shm_atime
;
1324 old_time32_t shm_dtime
;
1325 old_time32_t shm_ctime
;
1326 compat_ipc_pid_t shm_cpid
;
1327 compat_ipc_pid_t shm_lpid
;
1328 unsigned short shm_nattch
;
1329 unsigned short shm_unused
;
1330 compat_uptr_t shm_unused2
;
1331 compat_uptr_t shm_unused3
;
1334 struct compat_shminfo64
{
1335 compat_ulong_t shmmax
;
1336 compat_ulong_t shmmin
;
1337 compat_ulong_t shmmni
;
1338 compat_ulong_t shmseg
;
1339 compat_ulong_t shmall
;
1340 compat_ulong_t __unused1
;
1341 compat_ulong_t __unused2
;
1342 compat_ulong_t __unused3
;
1343 compat_ulong_t __unused4
;
1346 struct compat_shm_info
{
1347 compat_int_t used_ids
;
1348 compat_ulong_t shm_tot
, shm_rss
, shm_swp
;
1349 compat_ulong_t swap_attempts
, swap_successes
;
1352 static int copy_compat_shminfo_to_user(void __user
*buf
, struct shminfo64
*in
,
1355 if (in
->shmmax
> INT_MAX
)
1356 in
->shmmax
= INT_MAX
;
1357 if (version
== IPC_64
) {
1358 struct compat_shminfo64 info
;
1359 memset(&info
, 0, sizeof(info
));
1360 info
.shmmax
= in
->shmmax
;
1361 info
.shmmin
= in
->shmmin
;
1362 info
.shmmni
= in
->shmmni
;
1363 info
.shmseg
= in
->shmseg
;
1364 info
.shmall
= in
->shmall
;
1365 return copy_to_user(buf
, &info
, sizeof(info
));
1367 struct shminfo info
;
1368 memset(&info
, 0, sizeof(info
));
1369 info
.shmmax
= in
->shmmax
;
1370 info
.shmmin
= in
->shmmin
;
1371 info
.shmmni
= in
->shmmni
;
1372 info
.shmseg
= in
->shmseg
;
1373 info
.shmall
= in
->shmall
;
1374 return copy_to_user(buf
, &info
, sizeof(info
));
1378 static int put_compat_shm_info(struct shm_info
*ip
,
1379 struct compat_shm_info __user
*uip
)
1381 struct compat_shm_info info
;
1383 memset(&info
, 0, sizeof(info
));
1384 info
.used_ids
= ip
->used_ids
;
1385 info
.shm_tot
= ip
->shm_tot
;
1386 info
.shm_rss
= ip
->shm_rss
;
1387 info
.shm_swp
= ip
->shm_swp
;
1388 info
.swap_attempts
= ip
->swap_attempts
;
1389 info
.swap_successes
= ip
->swap_successes
;
1390 return copy_to_user(uip
, &info
, sizeof(info
));
1393 static int copy_compat_shmid_to_user(void __user
*buf
, struct shmid64_ds
*in
,
1396 if (version
== IPC_64
) {
1397 struct compat_shmid64_ds v
;
1398 memset(&v
, 0, sizeof(v
));
1399 to_compat_ipc64_perm(&v
.shm_perm
, &in
->shm_perm
);
1400 v
.shm_atime
= lower_32_bits(in
->shm_atime
);
1401 v
.shm_atime_high
= upper_32_bits(in
->shm_atime
);
1402 v
.shm_dtime
= lower_32_bits(in
->shm_dtime
);
1403 v
.shm_dtime_high
= upper_32_bits(in
->shm_dtime
);
1404 v
.shm_ctime
= lower_32_bits(in
->shm_ctime
);
1405 v
.shm_ctime_high
= upper_32_bits(in
->shm_ctime
);
1406 v
.shm_segsz
= in
->shm_segsz
;
1407 v
.shm_nattch
= in
->shm_nattch
;
1408 v
.shm_cpid
= in
->shm_cpid
;
1409 v
.shm_lpid
= in
->shm_lpid
;
1410 return copy_to_user(buf
, &v
, sizeof(v
));
1412 struct compat_shmid_ds v
;
1413 memset(&v
, 0, sizeof(v
));
1414 to_compat_ipc_perm(&v
.shm_perm
, &in
->shm_perm
);
1415 v
.shm_perm
.key
= in
->shm_perm
.key
;
1416 v
.shm_atime
= in
->shm_atime
;
1417 v
.shm_dtime
= in
->shm_dtime
;
1418 v
.shm_ctime
= in
->shm_ctime
;
1419 v
.shm_segsz
= in
->shm_segsz
;
1420 v
.shm_nattch
= in
->shm_nattch
;
1421 v
.shm_cpid
= in
->shm_cpid
;
1422 v
.shm_lpid
= in
->shm_lpid
;
1423 return copy_to_user(buf
, &v
, sizeof(v
));
1427 static int copy_compat_shmid_from_user(struct shmid64_ds
*out
, void __user
*buf
,
1430 memset(out
, 0, sizeof(*out
));
1431 if (version
== IPC_64
) {
1432 struct compat_shmid64_ds __user
*p
= buf
;
1433 return get_compat_ipc64_perm(&out
->shm_perm
, &p
->shm_perm
);
1435 struct compat_shmid_ds __user
*p
= buf
;
1436 return get_compat_ipc_perm(&out
->shm_perm
, &p
->shm_perm
);
1440 static long compat_ksys_shmctl(int shmid
, int cmd
, void __user
*uptr
, int version
)
1442 struct ipc_namespace
*ns
;
1443 struct shmid64_ds sem64
;
1446 ns
= current
->nsproxy
->ipc_ns
;
1448 if (cmd
< 0 || shmid
< 0)
1453 struct shminfo64 shminfo
;
1454 err
= shmctl_ipc_info(ns
, &shminfo
);
1457 if (copy_compat_shminfo_to_user(uptr
, &shminfo
, version
))
1462 struct shm_info shm_info
;
1463 err
= shmctl_shm_info(ns
, &shm_info
);
1466 if (put_compat_shm_info(&shm_info
, uptr
))
1473 err
= shmctl_stat(ns
, shmid
, cmd
, &sem64
);
1476 if (copy_compat_shmid_to_user(uptr
, &sem64
, version
))
1481 if (copy_compat_shmid_from_user(&sem64
, uptr
, version
))
1485 return shmctl_down(ns
, shmid
, cmd
, &sem64
);
1488 return shmctl_do_lock(ns
, shmid
, cmd
);
1495 COMPAT_SYSCALL_DEFINE3(shmctl
, int, shmid
, int, cmd
, void __user
*, uptr
)
1497 return compat_ksys_shmctl(shmid
, cmd
, uptr
, IPC_64
);
1500 #ifdef CONFIG_ARCH_WANT_COMPAT_IPC_PARSE_VERSION
1501 long compat_ksys_old_shmctl(int shmid
, int cmd
, void __user
*uptr
)
1503 int version
= compat_ipc_parse_version(&cmd
);
1505 return compat_ksys_shmctl(shmid
, cmd
, uptr
, version
);
1508 COMPAT_SYSCALL_DEFINE3(old_shmctl
, int, shmid
, int, cmd
, void __user
*, uptr
)
1510 return compat_ksys_old_shmctl(shmid
, cmd
, uptr
);
1516 * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
1518 * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
1519 * "raddr" thing points to kernel space, and there has to be a wrapper around
1522 long do_shmat(int shmid
, char __user
*shmaddr
, int shmflg
,
1523 ulong
*raddr
, unsigned long shmlba
)
1525 struct shmid_kernel
*shp
;
1526 unsigned long addr
= (unsigned long)shmaddr
;
1528 struct file
*file
, *base
;
1530 unsigned long flags
= MAP_SHARED
;
1533 struct ipc_namespace
*ns
;
1534 struct shm_file_data
*sfd
;
1536 unsigned long populate
= 0;
1543 if (addr
& (shmlba
- 1)) {
1544 if (shmflg
& SHM_RND
) {
1545 addr
&= ~(shmlba
- 1); /* round down */
1548 * Ensure that the round-down is non-nil
1549 * when remapping. This can happen for
1550 * cases when addr < shmlba.
1552 if (!addr
&& (shmflg
& SHM_REMAP
))
1555 #ifndef __ARCH_FORCE_SHMLBA
1556 if (addr
& ~PAGE_MASK
)
1562 } else if ((shmflg
& SHM_REMAP
))
1565 if (shmflg
& SHM_RDONLY
) {
1570 prot
= PROT_READ
| PROT_WRITE
;
1571 acc_mode
= S_IRUGO
| S_IWUGO
;
1574 if (shmflg
& SHM_EXEC
) {
1576 acc_mode
|= S_IXUGO
;
1580 * We cannot rely on the fs check since SYSV IPC does have an
1581 * additional creator id...
1583 ns
= current
->nsproxy
->ipc_ns
;
1585 shp
= shm_obtain_object_check(ns
, shmid
);
1592 if (ipcperms(ns
, &shp
->shm_perm
, acc_mode
))
1595 err
= security_shm_shmat(&shp
->shm_perm
, shmaddr
, shmflg
);
1599 ipc_lock_object(&shp
->shm_perm
);
1601 /* check if shm_destroy() is tearing down shp */
1602 if (!ipc_valid_object(&shp
->shm_perm
)) {
1603 ipc_unlock_object(&shp
->shm_perm
);
1609 * We need to take a reference to the real shm file to prevent the
1610 * pointer from becoming stale in cases where the lifetime of the outer
1611 * file extends beyond that of the shm segment. It's not usually
1612 * possible, but it can happen during remap_file_pages() emulation as
1613 * that unmaps the memory, then does ->mmap() via file reference only.
1614 * We'll deny the ->mmap() if the shm segment was since removed, but to
1615 * detect shm ID reuse we need to compare the file pointers.
1617 base
= get_file(shp
->shm_file
);
1619 size
= i_size_read(file_inode(base
));
1620 ipc_unlock_object(&shp
->shm_perm
);
1624 sfd
= kzalloc(sizeof(*sfd
), GFP_KERNEL
);
1630 file
= alloc_file_clone(base
, f_flags
,
1631 is_file_hugepages(base
) ?
1632 &shm_file_operations_huge
:
1633 &shm_file_operations
);
1634 err
= PTR_ERR(file
);
1641 sfd
->id
= shp
->shm_perm
.id
;
1642 sfd
->ns
= get_ipc_ns(ns
);
1645 file
->private_data
= sfd
;
1647 err
= security_mmap_file(file
, prot
, flags
);
1651 if (mmap_write_lock_killable(current
->mm
)) {
1656 if (addr
&& !(shmflg
& SHM_REMAP
)) {
1658 if (addr
+ size
< addr
)
1661 if (find_vma_intersection(current
->mm
, addr
, addr
+ size
))
1665 addr
= do_mmap(file
, addr
, size
, prot
, flags
, 0, &populate
, NULL
);
1668 if (IS_ERR_VALUE(addr
))
1671 mmap_write_unlock(current
->mm
);
1673 mm_populate(addr
, populate
);
1679 down_write(&shm_ids(ns
).rwsem
);
1680 shp
= shm_lock(ns
, shmid
);
1683 if (shm_may_destroy(shp
))
1684 shm_destroy(ns
, shp
);
1687 up_write(&shm_ids(ns
).rwsem
);
1696 SYSCALL_DEFINE3(shmat
, int, shmid
, char __user
*, shmaddr
, int, shmflg
)
1701 err
= do_shmat(shmid
, shmaddr
, shmflg
, &ret
, SHMLBA
);
1704 force_successful_syscall_return();
1708 #ifdef CONFIG_COMPAT
1710 #ifndef COMPAT_SHMLBA
1711 #define COMPAT_SHMLBA SHMLBA
1714 COMPAT_SYSCALL_DEFINE3(shmat
, int, shmid
, compat_uptr_t
, shmaddr
, int, shmflg
)
1719 err
= do_shmat(shmid
, compat_ptr(shmaddr
), shmflg
, &ret
, COMPAT_SHMLBA
);
1722 force_successful_syscall_return();
1728 * detach and kill segment if marked destroyed.
1729 * The work is done in shm_close.
1731 long ksys_shmdt(char __user
*shmaddr
)
1733 struct mm_struct
*mm
= current
->mm
;
1734 struct vm_area_struct
*vma
;
1735 unsigned long addr
= (unsigned long)shmaddr
;
1736 int retval
= -EINVAL
;
1740 VMA_ITERATOR(vmi
, mm
, addr
);
1743 if (addr
& ~PAGE_MASK
)
1746 if (mmap_write_lock_killable(mm
))
1750 * This function tries to be smart and unmap shm segments that
1751 * were modified by partial mlock or munmap calls:
1752 * - It first determines the size of the shm segment that should be
1753 * unmapped: It searches for a vma that is backed by shm and that
1754 * started at address shmaddr. It records it's size and then unmaps
1756 * - Then it unmaps all shm vmas that started at shmaddr and that
1757 * are within the initially determined size and that are from the
1758 * same shm segment from which we determined the size.
1759 * Errors from do_munmap are ignored: the function only fails if
1760 * it's called with invalid parameters or if it's called to unmap
1761 * a part of a vma. Both calls in this function are for full vmas,
1762 * the parameters are directly copied from the vma itself and always
1763 * valid - therefore do_munmap cannot fail. (famous last words?)
1766 * If it had been mremap()'d, the starting address would not
1767 * match the usual checks anyway. So assume all vma's are
1768 * above the starting address given.
1772 for_each_vma(vmi
, vma
) {
1774 * Check if the starting address would match, i.e. it's
1775 * a fragment created by mprotect() and/or munmap(), or it
1776 * otherwise it starts at this address with no hassles.
1778 if ((vma
->vm_ops
== &shm_vm_ops
) &&
1779 (vma
->vm_start
- addr
)/PAGE_SIZE
== vma
->vm_pgoff
) {
1782 * Record the file of the shm segment being
1783 * unmapped. With mremap(), someone could place
1784 * page from another segment but with equal offsets
1785 * in the range we are unmapping.
1787 file
= vma
->vm_file
;
1788 size
= i_size_read(file_inode(vma
->vm_file
));
1789 do_vma_munmap(&vmi
, vma
, vma
->vm_start
, vma
->vm_end
,
1792 * We discovered the size of the shm segment, so
1793 * break out of here and fall through to the next
1794 * loop that uses the size information to stop
1795 * searching for matching vma's.
1798 vma
= vma_next(&vmi
);
1804 * We need look no further than the maximum address a fragment
1805 * could possibly have landed at. Also cast things to loff_t to
1806 * prevent overflows and make comparisons vs. equal-width types.
1808 size
= PAGE_ALIGN(size
);
1809 while (vma
&& (loff_t
)(vma
->vm_end
- addr
) <= size
) {
1810 /* finding a matching vma now does not alter retval */
1811 if ((vma
->vm_ops
== &shm_vm_ops
) &&
1812 ((vma
->vm_start
- addr
)/PAGE_SIZE
== vma
->vm_pgoff
) &&
1813 (vma
->vm_file
== file
)) {
1814 do_vma_munmap(&vmi
, vma
, vma
->vm_start
, vma
->vm_end
,
1818 vma
= vma_next(&vmi
);
1821 #else /* CONFIG_MMU */
1822 vma
= vma_lookup(mm
, addr
);
1823 /* under NOMMU conditions, the exact address to be destroyed must be
1826 if (vma
&& vma
->vm_start
== addr
&& vma
->vm_ops
== &shm_vm_ops
) {
1827 do_munmap(mm
, vma
->vm_start
, vma
->vm_end
- vma
->vm_start
, NULL
);
1833 mmap_write_unlock(mm
);
1837 SYSCALL_DEFINE1(shmdt
, char __user
*, shmaddr
)
1839 return ksys_shmdt(shmaddr
);
1842 #ifdef CONFIG_PROC_FS
1843 static int sysvipc_shm_proc_show(struct seq_file
*s
, void *it
)
1845 struct pid_namespace
*pid_ns
= ipc_seq_pid_ns(s
);
1846 struct user_namespace
*user_ns
= seq_user_ns(s
);
1847 struct kern_ipc_perm
*ipcp
= it
;
1848 struct shmid_kernel
*shp
;
1849 unsigned long rss
= 0, swp
= 0;
1851 shp
= container_of(ipcp
, struct shmid_kernel
, shm_perm
);
1852 shm_add_rss_swap(shp
, &rss
, &swp
);
1854 #if BITS_PER_LONG <= 32
1855 #define SIZE_SPEC "%10lu"
1857 #define SIZE_SPEC "%21lu"
1861 "%10d %10d %4o " SIZE_SPEC
" %5u %5u "
1862 "%5lu %5u %5u %5u %5u %10llu %10llu %10llu "
1863 SIZE_SPEC
" " SIZE_SPEC
"\n",
1868 pid_nr_ns(shp
->shm_cprid
, pid_ns
),
1869 pid_nr_ns(shp
->shm_lprid
, pid_ns
),
1871 from_kuid_munged(user_ns
, shp
->shm_perm
.uid
),
1872 from_kgid_munged(user_ns
, shp
->shm_perm
.gid
),
1873 from_kuid_munged(user_ns
, shp
->shm_perm
.cuid
),
1874 from_kgid_munged(user_ns
, shp
->shm_perm
.cgid
),